Conventional short range data radios typically divide the radio spectrum within which they operate into non-overlapping frequency channels. For example, radios with a 1 Megahertz (MHz) occupied bandwidth operating in the 2.4 Gigahertz (GHz) Industrial Scientific Medical (ISM) band typically divide that spectrum into approximately 80 1-MHz wide channels. These radio systems transmit and receive data using frequency modulated Radio Frequency (RF) signals centered on one of these 1-MHz channels. In some cases, the transmitter may hop between channels during normal data transmission. In other cases, having found a good channel, the transmitter may continue to use that one channel unless or until data transfer on that channel becomes unreliable.
Typically, these radio systems generate the RF carrier frequency by multiplying the frequency of a low frequency crystal oscillator up to the RF frequency used for transmission. Many 2.4 GHz radio systems use 13 MHz crystals for this purpose, but crystal frequencies in the 12-32 MHz range are also common.
Radio receivers, especially Frequency Modulation (FM) receivers using a low Intermediate Frequency (IF), typically implement a Band Pass Filter (BPF), through which the mixed-down signal is passed before demodulation. This is necessary in order to prevent RF signals on adjacent channels from being demodulated or interfering with the reception of signals on the channel the receiver is currently configured to receive.
The crystal oscillators of both the transmitter and receiver should be oscillating at almost exactly the same frequency. If not, part of the transmitted signal may be attenuated by the receiver's BPF. Frequency response 12 in FIG. 1A shows the frequency spectrum of a transmit signal 18 and the frequency operation of a receiver's Band Pass Filter (BPF) 16. When the transmitter and receiver have crystal oscillators with the same frequency, the transmit signal 18 should be substantially centered within the BPF 16. In this common reference frequency condition, the transmitted signal 18 will have minimum attenuation.
Frequency response 14 in FIG. 1B shows the frequency spectrum when the transmitter and receiver have crystal oscillators with different (offset) reference frequencies. In this offset frequency situation, the transmit signal 18 is no longer centered within the BPF 16. Any portion of the transmit signal 18 extending outside of BPF 16 is attenuated, such as the shaded portion 20. The attenuation 20 lowers the signal strength of signal 18 and can prevent the receiver from successfully or reliably receiving data carried in the transmit signal 18.
In one example, a radio system may operate at 2.4 GHz, with a 1 MHz channel spacing and a 900 kHz occupied bandwidth. At 2450 MHz, a 50 parts per million (ppm) offset is equal to 122.5 kHz. As the occupied bandwidth is 100 kHz less than the channel spacing, there is 50 kHz on either side of a perfectly centered transmitter spectrum that is not part of the adjacent channel. In the 50 ppm offset example, 7.2% of the transmitted signal extends into the adjacent channel. Typically, the receiver BPF is a little wider than the channel, and the roll-off of the filter is not a “brick wall”, so a small offset can be tolerated with minimal impact on receive sensitivity. In a typical 2.4 GHz radio system, a 50 ppm offset is approximately the maximum that can be tolerated without significantly impacting performance.
Conventional wireless solutions use quartz crystals to derive a radio carrier reference frequency. These conventional solutions have disadvantages, including requiring expensive, high accuracy crystals. Even using such crystals, significant offsets may exist between the transmit and receive frequencies resulting in reduced receive sensitivity. Moderately affordable crystals may require time-of-manufacture crystal trimming, thereby increasing manufacturing cost and complexity.
The receiver and transmitter are each subject to separate oscillator frequency inaccuracies. Therefore, the receiver and transmitter require a crystal with an accuracy of better than ±25 ppm to prevent the oscillator accuracy from impacting system performance with a combined worst case error of greater than 50 ppm.
Crystal oscillator accuracy is typically specified as three components: initial tolerance, temperature variation, and long-term drift. In order to put products in the best possible light, crystal oscillator vendors typically quote only the initial tolerance. Frequency variation with temperature is usually similar to the initial tolerance, and aging is usually in the range of one to five ppm per year.
Crystals generally drift in the same direction. Two instances of the same crystal would not usually drift in opposite directions, but they may well drift at different rates in the same direction. One factor affecting drift is the drive strength of the oscillator circuit driving the crystal. Another factor may be the proportion of time that the crystal oscillator is active. Over 5 years, a crystal with a 3 ppm/year drift spec may drift only 5 ppm (or less) or not at all, while another may drift 15 ppm. This crystal drift depends on drive strength, the amount of time that the oscillator is running, and the physical properties of the individual crystal.
The frequency variation with temperature is not linear, but rather typically a quadratic or cubic curve. Therefore, variations in temperature across only part of the rated range may cause frequency to vary over most of the stated tolerance.
Therefore, a 25 ppm crystal, which initially may appear to be suitable for uses in the wireless applications discussed above, may not in fact be suitable. Such a crystal would typically have 25 ppm initial tolerance, 25 ppm variation over temperature, and 3 ppm/year drift. One such crystal, starting at −25 ppm, and operated at a temperature which caused the frequency to oscillate at close to its minimum frequency, may be oscillating at −50 ppm from its nominal after 5 years. Another crystal, starting at +25 ppm, at a different temperature may oscillate at +65 ppm from its nominal after 5 years, resulting in a difference of 115 ppm. This drift could severely impact the receive sensitivity of almost any 2.4 GHz radio system. Therefore, to meet the ±25 ppm specification discussed above, a more expensive 10 ppm crystal is required.
Above 30 ppm, the cost savings from specifying a looser tolerance is low. For example, one vendor may offer a 13 MHz 30 ppm crystal in volume at a given price, and the 50 ppm version of the same crystal may only be a few cents cheaper. However, crystals with tolerances below 30 ppm quickly become more expensive, and a 10 ppm crystal may typically cost 3× to 5× more than the 30 ppm crystal. Frequency accuracy requirements of a design may therefore place a significant cost burden on low cost wireless systems.
For this reason, many low cost radio Integrated Circuits (ICs) include a feature allowing trimming of the initial crystal frequency. Typically, this is implemented by using a digitally trimmable capacitance. At manufacturing test of a wireless product, the oscillator frequency is measured, and an appropriate trim factor is stored in non-volatile memory within the device. This trim factor is loaded into a radio Integrated Circuit (IC) after each reset and allows the digitally controlled capacitance of the crystal oscillator to tune the crystal frequency to a nominal value. This removes the initial tolerance component of the oscillator, reducing the variation to just the temperature and drift components. This technique allows the use of 15 ppm crystals with many 2.4 GHz radio ICs, without impacting radio performance. However, this comes at the cost and trouble of implementing crystal tuning during manufacture.
It would be desirable to use much less accurate crystals in low cost radio transmitters and receivers without the need of a factory calibration or memory to store said calibration. It would also be desirable to compensate for reference frequency offsets between transmitters and receivers due to variables such as temperature and power supply voltage.